Driftproof

tao-generate-image-embeddings: worth installing? Reported results

Reported results for tao-generate-image-embeddings (NVIDIA/skills) from NVIDIA, one row per receipt, in plain words.

Skill skills/tao-generate-image-embeddings in NVIDIA/skills.

One row per receipt. No figure on this page combines two rows.

Reported by
Reported by NVIDIA in its benchmark card
Model
Claude Opus 4.8 (aws/anthropic/bedrock-claude-opus-4-8, as the source names it)
Harness
claude-code, version not stated
Date
2026-09-17
Tasks and runs
1 task, as printed
Result
NVIDIA reported a score of 99.6% overall and no uplift figure (3 attempts per task). Overall: 99.6% — baseline ran, but no comparable score was available; uplift unavailable Security: 100.0% → 100.0% (±0.0 points) Correctness: 13.3% → 100.0% (+86.7 points) Discoverability: 100.0% — baseline ran, but no comparable score was available; uplift unavailable Effectiveness: 10.0% → 100.0% (+90.0 points) Efficiency: 98.0% — baseline ran, but no comparable score was available; uplift unavailable
Tokens
65,925 tokens with the skill and 119,856 without, over all cases, as printed; the card gives no cost
Freshness
Reported on Opus 4.8. Registered opus models released after this run, as of the build date: Opus 5.5.
Skill commit
14a98ae278cec7e867695189d0f4003c08e4f765
Receipt
tao-generate-image-embeddings-aws-anthropic-bedrock-claude-opus-4-8-imported-nvidia-cards-claude-code-2026-09-17-ff550d1ef72b.json
Reported by
Reported by NVIDIA in its benchmark card
Model
GPT-5.5 (openai/openai/gpt-5.5, as the source names it)
Harness
codex, version not stated
Date
2026-09-17
Tasks and runs
1 task, as printed
Result
NVIDIA reported a score of 71.7% overall and no uplift figure (3 attempts per task). Overall: 71.7% — baseline ran, but no comparable score was available; uplift unavailable Security: 100.0% → 100.0% (±0.0 points) Correctness: 30.0% → 100.0% (+70.0 points) Discoverability: 0.0% — baseline ran, but no comparable score was available; uplift unavailable Effectiveness: 22.5% → 60.0% (+37.5 points) Efficiency: 99.7% → 98.5% (-1.2 points)
Tokens
13,947 tokens with the skill and 27,095 without, over all cases, as printed; the card gives no cost
Freshness
Reported on GPT-5.5. Freshness unknown: the registry has no release date for it.
Skill commit
14a98ae278cec7e867695189d0f4003c08e4f765
Receipt
tao-generate-image-embeddings-openai-openai-gpt-5-5-imported-nvidia-cards-codex-2026-09-17-ff550d1ef72b.json

Figures from NVIDIA/skills at commit 14a98ae278cec7e867695189d0f4003c08e4f765, licensed CC BY 4.0. Reformatted by Driftproof; NVIDIA does not endorse this page.

Disagree, or want a re-run?

Built on from committed receipts.